Skip to content

sensing·

⏱ 45s

TINY sees through the head camera and can inject what it sees straight into the conversation. All sensing is read-only.

reachy_camera — grab a frame·

reachy_camera(save_path="", timeout=10.0)   # → saves a JPEG, returns the path

Captures a single frame from the head camera via the daemon. With no save_path, it uses TINY_CAMERA_SNAPSHOT (default /tmp/tiny_view.jpg).

reachy_look_at — visual servoing·

reachy_look_at(u, v, duration=1.0)   # pixel coords in the camera frame

Points the head so pixel (u,v) moves toward center. Chain it after a detection to track a face or object.

take_photo — bidi vision (the good one)·

take_photo(question="who's there?")

Grabs a frame and injects it into the model's context with your question — so TINY actually reasons about what it sees and replies aloud. This is the tool to use for "look at me", "what do you see?", "who's there?".

"look at me"        → take_photo(question="who is in front of me?")
"what's this?"      → take_photo(question="what object am I looking at?")
"read this to me"   → take_photo(question="transcribe any text you see")

Media backend

The daemon owns the camera. If frame grabs fight the daemon, set REACHY_MEDIA_BACKEND=no_media and let reachy_camera grab transiently, or REACHY_CAMERA_BACKEND=local. See env vars.